If you're reading this book, chances are pretty good that you are already a game developer. That being the case, then you already know just how exciting it is to program your own games, either professionally or as a highly gratifying hobby that is very time-consuming. Now you're ready to take your game programming skills to the next level—that is, you're ready to implement multiplayer functionality into your JavaScript-based games.
In case you have already set out to create multiplayer games for the Open Web Platform using HTML5 and JavaScript, then you may have already come to realize that a personal desktop computer, laptop, or a mobile device is not particularly the most appropriate device to share with another human player for games in which two or more players share the same game world at the same time. Therefore, what is needed in order to create exciting multiplayer games with JavaScript is some form of networking technology.
In this chapter, we will discuss the following principles and concepts:
- The basics of networking and network programming paradigms
- Socket programming with HTML5
- Programming a game server and game clients
- Turn-based multiplayer games
It is said that one cannot program games that make use of networking without first understanding all about the discipline of computer networking and network programming. Although having a deep understanding of any topic can be only beneficial to the person working on that topic, I don't believe that you must know everything there is to know about game networking in order to program some pretty fun and engaging multiplayer games. Saying that is the case is like saying that one needs to be a scholar of the Spanish language in order to cook a simple burrito. Thus, let us take a look at the most basic and fundamental concepts of networking. At the end of this section, you will know enough about computer networking to get started, and you will feel comfortable adding multiplayer aspects to your games.
One thing to keep in mind is that, even though networked games are not nearly as old as single-player games, computer networking is actually a very old and well-studied subject. Some of the earliest computer network systems date back to the 1950s. Though some of the techniques have improved over the years, the basic idea remains the same: two or more computers are connected together to establish communication between the machines. By communication, I mean data exchange, such as sending messages back and forth between the machines, or one of the machines only sends the data and the other only receives it.
With this brief introduction to the concept of networking, you are now grounded in the subject of networking, enough to know what is required to network your games—two or more computers that talk to each other as close to real time as possible.
By now, it should be clear how this simple concept makes it possible for us to connect multiple players into the same game world. In essence, we need a way to share the global game data among all the players who are connected to the game session, then continue to update each player about every other player. There are several different techniques that are commonly used to achieve this, but the two most common approaches are peer-to-peer and client-server. Both techniques present different opportunities, including advantages and disadvantages. In general, neither is particularly better than the other, but different situations and use cases may be better suited for one or the other technique.
A simple way to connect players into the same virtual game world is through the peer-to-peer architecture. Although the name might suggest that only two peers ("nodes") are involved, by definition a peer-to-peer network system is one in which two or more nodes are connected directly to each other without a centralized system orchestrating the connection or information exchange.
On a typical peer-to-peer setup, each peer serves the same function as every other one—that is, they all consume the same data and share whatever data they produce so that others can stay synchronized. In the case of a peer-to-peer game, we can illustrate this architecture with a simple game of Tic-tac-toe.
Once both the players have established a connection between themselves, whoever is starting the game makes a move by marking a cell on the game board. This information is relayed across the wire to the other peer, who is now aware of the decision made by his or her opponent, and can thus update their own game world. Once the second player receives the game's latest state that results from the first player's latest move, the second player is able to make a move of their own by checking some available space on the board. This information is then copied over to the first player who can update their own world and continue the process by making the next desired move.
The process goes on until one of the peers disconnects or the game ends as some condition that is based on the game's own business logic is met. In the case of the game of Tic-tac-toe, the game would end once one of the players has marked three spaces on the board forming a straight line or if all nine cells are filled, but neither player managed to connect three cells in a straight path.
Some of the benefits of peer-to-peer networked games are as follows:
- Fast data transmission: Here, the data goes directly to its intended target. In other architectures, the data could go to some centralized node first, then the central node (or the "server", as we'll see in the next section) contacts the other peer, sending the necessary updates.
- Simpler setup: You would only need to think about one instance of your game that, generally speaking, handles its own input, sends its input to other connected peers, and handles their output as input for its own system. This can be especially handy in turn-based games, for example, most board games such as Tic-tac-toe.
- More reliability: Here one peer that goes offline typically won't affect any of the other peers. However, in the simple case of a two-player game, if one of the players is unable to continue, the game will likely cease to be playable. Imagine, though, that the game in question has dozens or hundreds of connected peers. If a handful of them suddenly lose their Internet connection, the others can continue to play. However, if there is a server that is connecting all the nodes and the server goes down, then none of the other players will know how to talk to each other, and nobody will know what is going on.
On the other hand, some of the more obvious drawbacks of peer-to-peer architecture are as follows:
- Incoming data cannot be trusted: Here, you don't know for sure whether or not the sender modified the data. The data that is input into a game server will also suffer from the same challenge, but once the data is validated and broadcasted to all the other peers, you can be more confident that the data received by each peer from the server will have at least been sanitized and verified, and will be more credible.
- Fault tolerance can be very low: The opposite argument was made in the benefits' section of Peer-to-peer networking that we discussed previously; if enough players share the game world, one or more crashes won't make the game unplayable to the rest of the peers. Now, if we consider the many cases where any of the players that suddenly crash out of the game negatively affect the rest of the players, we can see how a server could easily recover from the crash.
- Data duplication when broadcasting to other peers: Imagine that your game is a simple 2D side scroller, and many other players are sharing that game world with you. Every time one of the players moves to the right, you receive the new (x, y) coordinates from that player, and you're able to update your own game world. Now, imagine that you move your player to the right by a very few pixels; you would have to send that data out to all of the other nodes in the system.
Overall, peer-to-peer is a very powerful networking architecture and is still widely used by many games in the industry. Since current peer-to-peer web technologies are still in their infancy, most JavaScript-powered games today do not make use of peer-to-peer networking. For this and other reasons that should become apparent soon, we will focus the rest of the book almost exclusively on the other popular networking paradigm, namely, the client-server architecture.
The idea behind the client-server networking architecture is very simple. If you squint your eyes hard enough, you can almost see a peer-to-peer graph. The most obvious difference between them, is that, instead of every node being an equal peer, one of the nodes is special. That is, instead of every node connecting to every other node, every node (client) connects to a main centralized node called the server.
While the concept of a client-server network seems clear enough, perhaps a simple metaphor might make it easier for you to understand the role of each type of node in this network format as well as differentiate it from peer-to-peer (McConnell, Steve, (2004) Code Complete., Microsoft Press). In a peer-to-peer network, you can think of it as a group of friends (peers) having a conversation at a party. They all have access to all the other peers involved in the conversation and can talk to them directly. On the other hand, a client-server network can be viewed as a group of friends having dinner at a restaurant. If a client of the restaurant wishes to order a certain item from the menu, he or she must talk to the waiter, who is the only person in that group of people with access to the desired products and the ability to serve the products to the clients.
In short, the server is in charge of providing data and services to one or more clients. In the context of game development, the most common scenario is when two or more clients connect to the same server; the server will keep track of the game as well as the distributed players. Thus, if two players are to exchange information that is only pertinent to the two of them, the communication will go from the first player to and through the server and will end up at the other end with the second player.
Following the example of the two players involved in a game of Tic-tac-toe that we looked at in the section about peer-to-peer, we can see how similar the flow of events is on a client-server model. Again, the main difference is that players are unaware of each other and only know what the server tells them.
While you can very easily mimic a peer-to-peer model by using a server to merely connect the two players, most often the server is used much more actively than that. There are two ways to engage the server in a networked game, namely in an authoritative and a non-authoritative way. That is to say, you can have the enforcement of the game's logic strictly in the server, or you can have the clients handle the game logic, input validation, and so on. Today, most games using the client-server architecture actually use a hybrid of the two (authoritative and non-authoritative servers, which we'll discuss later in the book). For all intents and purposes, however, the server's purpose in life is to receive input from each of the clients and distribute that input throughout the pool of connected clients.
Now, regardless of whether you decide to go with an authoritative server instead of a non-authoritative one, you will notice that one of challenges with a client-server game is that you will need to program both ends of the stack. You will have to do this even if your clients do nothing more than take input from the user, forward it to the server, and render whatever data they receive from the server; if your game server does nothing more than forward the input that it receives from each client to every other client, you will still need to write a game client and a game server.
We will discuss game clients and servers later in the chapter. For now, all we really need to know is that these two components are what set this networking model apart from peer-to-peer.
Some of the benefits of client-server networked games are as follows:
- Separation of concerns: If you know anything about software development, you know that this is something you should always aim for. That is, good, maintainable software is written as discrete components where each does one "thing", and it is done well. Writing individual specialized components lets you focus on performing one individual task at a time, making your game easier to design, code, test, reason, and maintain.
- Centralization: While this can be argued against as well as in favor of, having one central place through which all communication must flow makes it easier to manage such communication, enforce any required rules, control access, and so forth.
- Less work for the client: Instead of having a client (peer) in charge of taking input from the user as well as other peers, validating all the input, sharing data among other peers, rendering the game, and so on, the client can focus on only doing a few of these things, allowing the server to offload some of this work. This is particularly handy when we talk about mobile gaming, and how much subtle divisions of labor can impact the overall player experience. For example, imagine a game where 10 players are engaged in the same game world. In a peer-to-peer setup, every time one player takes an action, he or she would need to send that action to nine other players (in other words, there would need to be nine network calls, boiling down to more mobile data usage). On the other hand, on a client-server configuration, one player would only need to send his or her action to one of the peers, that is, the server, who would then be responsible for sending that data to the remaining nine players.
Common drawbacks of client-server architectures, whether or not the server is authoritative, are as follows:
- Communication takes longer to propagate: In the very best possible scenario imaginable, every message sent from the first player to the second player would take twice as long to be delivered as compared to a peer-to-peer connection. That is, the message would be first sent from the first player to the server and then from the server to the second player. There are many techniques that are used today to solve the latency problem faced in this scenario, some of which we will discuss in much more depth later in Chapter 4, Reducing Network Latency. However, the underlying dilemma will always be there.
- More complexity due to more moving parts: It doesn't really matter how you slice the pizza; the more code you need to write (and trust me, when you build two separate modules for a game, you will write more code), the greater your mental model will have to be. While much of your code can be reused between the client and the server (especially if you use well-established programming techniques, such as object-oriented programming), at the end of the day, you need to manage a greater level of complexity.
- Single point of failure and network congestion: Up until now, we have mostly discussed the case where only a handful of players participate in the same game. However, the more common case is that a handful of groups of players play different games at the same time.
Using the same example of the two-player game of Tic-tac-toe, imagine that there are thousands of players facing each other in single games. In a peer-to-peer setup, once a couple of players have directly paired off, it is as though there are no other players enjoying that game. The only thing to keep these two players from continuing their game is their own connection with each other.
On the other hand, if the same thousands of players are connected to each other through a server sitting between the two, then two singled out players might notice severe delays between messages because the server is so busy handling all of the messages from and to all of the other people playing isolated games. Worse yet, these two players now need to worry about maintaining their own connection with each other through the server, but they also hope that the server's connection between them and their opponent will remain active.
All in all, many of the challenges involved in client-server networking are well studied and understood, and many of the problems you're likely to face during your multiplayer game development will already have been solved by someone else. Client-server is a very popular and powerful game networking model, and the required technology for it, which is available to us through HTML5 and JavaScript, is well developed and widely supported.
By discussing some of the ways in which your players can talk to each other across some form of network, we have yet only skimmed over how that communication is actually done. Let us then describe what protocols are and how they apply to networking and, more importantly, multiplayer game development.
The word protocol can be defined as a set of conventions or a detailed plan of a procedure [Citation [Def. 3,4]. (n.d.). In Merriam Webster Online, Retrieved February 12, 2015, from http://www.merriam-webster.com/dictionary/protocol]. In computer networking, a protocol describes to the receiver of a message how the data is organized so that it can be decoded. For example, imagine that you have a multiplayer beat 'em up game, and you want to tell the game server that your player just issued a kick command and moved 3 units to the left. What exactly do you send to the server? Do you send a string with a value of "kick", followed by the number 3? Otherwise, do you send the number first, followed by a capitalized letter "K", indicating that the action taken was a kick? The point I'm trying to make is that, without a well-understood and agreed-upon protocol, it is impossible to successfully and predictably communicate with another computer.
The two networking protocols that we'll discuss in the section, and that are also the two most widely used protocols in multiplayer networked games, are the Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP). Both protocols provide communication services between clients in a network system. In simple terms, they are protocols that allow us to send and receive packets of data in such a way that the data can be identified and interpreted in a predictable way.
When data is sent through TCP, the application running in the source machine first establishes a connection with the destination machine. Once a connection has been established, data is transmitted in packets in such a way that the receiving application can then put the data back together in the appropriate order. TCP also provides built-in error checking mechanisms so that, if a packet is lost, the target application can notify the sender application, and any missing packets are sent again until the entire message is received.
In short, TCP is a connection-based protocol that guarantees the delivery of the full data in the correct order. Use cases where this behavior is desirable are all around us. When you download a game from a web server, for example, you want to make sure that the data comes in correctly. You want to be sure that your game assets will be properly and completely downloaded before your users start playing your game. While this guarantee of delivery may sound very reassuring, it can also be thought of as a slow process, which, as we'll see briefly, may sometimes be more important than knowing that the data will arrive in full.
In contrast, UDP transmits packets of data (called datagrams) without the use of a pre-established connection. The main goal of the protocol is to be a very fast and frictionless way of sending data towards some target application. In essence, you can think of UDP as the brave employees who dress up as their company's mascot and stand outside their store waving a large banner in the hope that at least some of the people driving by will see them and give them their business.
While at first, UDP may seem like a reckless protocol, the use cases that make UDP so desirable and effective includes the many situations when you care more about speed than missing packets a few times, getting duplicate packets, or getting them out of order. You may also want to choose UDP over TCP when you don't care about the reply from the receiver. With TCP, whether or not you need some form of confirmation or reply from the receiver of your message, it will still take the time to reply back to you, at least acknowledging that the message was received. Sometimes, you may not care whether or not the server received the data.
A more concrete example of a scenario where UDP is a far better choice over TCP is when you need a heartbeat from the client letting the server know if the player is still there. If you need to let your server know that the session is still active every so often, and you don't care if one of the heartbeats get lost every now and again, then it would be wise to use UDP. In short, for any data that is not mission-critical and you can afford to lose, UDP might be the best option.
In closing, keep in mind that, just as peer-to-peer and client-server models can be built side by side, and in the same way your game server can be a hybrid of authoritative and non-authoritative, there is absolutely no reason why your multiplayer games should only use TCP or UDP. Use whichever protocol a particular situation calls for.
There is one other protocol that we'll cover very briefly, but only so that you can see the need for network sockets in game development. As a JavaScript programmer, you are doubtlessly familiar with Hypertext Transfer Protocol (HTTP). This is the protocol in the application layer that web browsers use to fetch your games from a Web server.
While HTTP is a great protocol to reliably retrieve documents from web servers, it was not designed to be used in real-time games; therefore, it is not ideal for this purpose. The way HTTP works is very simple: a client sends a request to a server, which then returns a response back to the client. The response includes a completion status code, indicating to the client that the request is either in process, needs to be forwarded to another address, or is finished successfully or erroneously (Hypertext Transfer Protocol (HTTP/1.1): Authentication, (June 1999). https://tools.ietf.org/html/rfc7235)
There are a handful of things to note about HTTP that will make it clear that a better protocol is needed for real-time communication between the client and server. Firstly, after each response is received by the requester, the connection is closed. Thus, before making each and every request, a new connection must be established with the server. Most of the time, an HTTP request will be sent through TCP, which, as we've seen, can be slow, relatively speaking.
Secondly, HTTP is by design a stateless protocol. This means that, every time you request a resource from a server, the server has no idea who you are and what is the context of the request. (It doesn't know whether this is your first request ever or if you're a frequent requester.) A common solution to this problem is to include a unique string with every HTTP request that the server keeps track of, and can thus provide information about each individual client on an ongoing basis. You may recognize this as a standard session. The major downside with this solution, at least with regard to real-time gaming, is that mapping a session cookie to the user's session takes additional time.
Finally, the major factor that makes HTTP unsuitable for multiplayer game programming is that the communication is one way—only the client can connect to the server, and the server replies back through the same connection. In other words, the game client can tell the game server that a punch command has been entered by the user, but the game server cannot pass that information along to other clients. Think of it like a vending machine. As a client of the machine, we can request specific items that we wish to buy. We formalize this request by inserting money into the vending machine, and then we press the appropriate button.
Under no circumstance will a vending machine issue commands to a person standing nearby. That would be like waiting for a vending machine to dispense food, expecting people to deposit the money inside it afterwards.
The answer to this lack of functionality in HTTP is pretty straightforward. A network socket is an endpoint in a connection that allows for two-way communication between the client and the server. Think of it more like a telephone call, rather than a vending machine. During a telephone call, either party can say whatever they want at any given time. Most importantly, the connection between both parties remains open throughout the duration of the conversation, making the communication process highly efficient.
WebSocket is a protocol built on top of TCP, allowing web-based applications to have two-way communication with a server (The WebSocket Protocol, (December 2011). http://tools.ietf.org/html/rfc6455 RFC 6455). The way a WebSocket is created consists of several steps, including a protocol upgrade from HTTP to WebSocket. Thankfully, all of the heavy lifting is done behind the scenes by the browser and JavaScript, as we'll see in the next section. For now, the key takeaway here is that with a TCP socket (yes, there are other types of socket including UDP sockets), we can reliably communicate with a server, and the server can talk back to us as per the need.